XScientist protocol makes AI research reproducible like Git

The paper, posted on arXiv by Jixiang Luo, describes XScientist as a solution to a critical blind spot in autonomous research systems. Current AI researchers typically evaluate systems as “one-shot paper generators” that produce a manuscript and sparse logs, making audit and reproduction nearly impossible. XScientist instead treats each run as a portable, git-like research artifact. It orchestrates every step—from idea generation and experiment execution to manuscript drafting, self-review, repair, and quality gating—as one continuously observable pipeline.

The protocol exports an Agent-Native Research Artifact (ARA) that records an exploration DAG (directed acyclic graph). Each node contains code, outputs, claim-to-evidence anchors, content hashes, provenance, and re-execution hooks. This design ensures that failed branches, repaired experiments, ablations, and manuscript claims remain connected to the nodes that produced them. XScientist also includes deterministic integrity forensics, sample gates, truth contracts, reviewer-oriented repair loops, and long-running daemon controls. The implementation is publicly available on GitHub, aiming to move autonomous science from single-run demos toward reproducible, reviewable, and forkable research infrastructure.